The screw classification task so far is only performed by humans. In fact, the photos are not accepted as input for the screw classification because there is information that cannot be determined with the images, such as the diameter of the screw and the pitch of the thread. With the advancement of machine learning models and the inclusion of automatic classification of digital images with deep neural network architectures, the solution to this task has not been explored, largely because the transcendental factor for its training is an appropriate data set that does not exist for this problem. In the present project, a set of unpublished images was built with which it is intended to train deep neural networks for the classification of screws. In addition, a specialized object detection model for screws was trained, which will work together with the classification model so that, apart from giving a classification, it identifies which part of the image the screw is in. Finally, the models were put into production within an interface in which the objective is to upload an image with screws and for the models to be able to detect where they are and classify their features